A potential source of bias in the data is early death due to smoking in early-onset smokers. This could artificially cause an interaction between rs16969968 and early smoking onset if the early deaths were preferentially in individuals who lack the risk allele. However, if this were the case, we would expect to see a stronger interaction among individuals born in the 1920–1939 birth cohort compared with in the 1960–1979 birth cohort, which is not the case.
